OpenAI starts embedded assessments with Apollo, METR, Transluce and Goodfire
Original title: Our first phase of embedded assessments
Assessors get internal checkpoints, code, data and monitoring access to test alignment claims and safeguard effectiveness.
OpenAI signed agreements with Apollo Research, METR, Transluce and Goodfire for a first phase of embedded assessments. Independent experts will use internal models, training artifacts and monitoring systems to check alignment and safeguards over an extended period. The work supports OpenAI’s commitments under the White House Accord on Super Intelligence. A broader METR monitor assessment is still pending agreement on scope.
Unconfirmed: A second, broader METR monitor assessment is only preliminarily agreed and still pending scope.
- Apollo will test four claims about covert model subversion and detection.
- Goodfire will monitor training runs and debug alignment with its interpretability platform.
- METR will review misaligned behavior in OpenAI’s largest frontier reinforcement-learning run.
- Transluce will examine quality and coverage of post-training alignment evaluations.
- Organizations
- OpenAIApollo ResearchMETRTransluceGoodfire
From the original article
10 min read at openai.comOpenAI begins embedded assessments with Apollo Research, METR, and Transluce, and Goodfire to independently evaluate model alignment, safeguards, and monitoring systems.
openai.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.